Failure Index / Rate limits & availability · warning · LangChain · OpenAI Agents SDK · CrewAI

Agent hits HTTP 429: rate limit exceeded (often from its own retry loop)

The agent is calling a service faster than allowed. The nasty version: an earlier failure sends the agent into a retry loop, and the retries THEMSELVES trigger the rate limit — the 429 you see is a symptom of a different bug upstream.

The error

agent 429 too many requests loop
langchain RateLimitError tool
openai agents 429 retry

Root cause

Either the workload genuinely exceeds the provider's limits, or (very common in agents) a failing step is being retried in a tight loop, multiplying the request volume. Parallel agent runs sharing one key also stack their quotas.

The fix

First check the steps BEFORE the 429 — if the same tool call repeats, fix that failure; the rate limit will disappear with it. If traffic is legitimate, add exponential backoff with jitter and respect the Retry-After header.

Preventing it next time

Cap agent retries (2-3 with backoff), and alert on repeated identical tool calls within one run — a loop detector catches this class before the provider does.

Related failures